Training large AI models has become one of the biggest challenges in modern computing—not just because of complexity, but because of cost, power use, and wasted resources. A new research paper from DeepSeek proposes an approach that could help ease some of those pressures.

The method, called manifold-constrained hyperconnection (mHC), focuses on making large AI models easier and more reliable to train. Instead of chasing raw performance gains, the idea is to reduce instability during training—a common issue that forces companies to restart expensive training runs from scratch.

In simple terms, many advanced AI models fail mid-training. When that happens, weeks of work, massive amounts of electricity, and thousands of GPU hours are lost. DeepSeek’s approach aims to prevent those failures by keeping model behavior more predictable, even as models grow larger.

This matters because AI training today consumes enormous amounts of power. While mHC doesn’t make GPUs themselves use less power, it can reduce wasted power by helping models finish training without crashing or needing repeated restarts.

Another benefit is efficiency at scale. When training is more stable, companies don’t need to rely as heavily on “brute force” methods—such as throwing more GPUs, more memory, or longer training schedules at a problem just to make it work. That can lower the total energy used over the full training process.

DeepSeek’s research doesn’t claim to solve hardware shortages or energy challenges overnight. Instead, it represents a quieter but important improvement: making better use of the resources already available. Over time, techniques like this could help AI developers train powerful models with fewer wasted compute hours and lower overall energy consumption.

As language models continue to grow, reducing inefficiency may become just as important as chasing higher performance—and that’s where DeepSeek’s new AI architecture could make a real difference.

Don’t miss a thing! Join our Telegram community for instant updates and grab our free daily newsletter for the best tech stories!

For more daily updates, please visit our News Section.

(Sources: 1, 2 | Image)

The post DeepSeek kicks off 2026 with new AI architecture aimed at more efficient model training appeared first on Gizmochina.

Highlights:

• Chinese open-source AI models have surged to nearly 30% of global usage, up from just 1% a year earlier.
  
• Major U.S. tech leaders and startups are adopting Chinese models due to lower costs, fast iteration, and strong performance.
  
• Nvidia CEO Jensen Huang warns that China’s rapid infrastructure growth and energy capacity give it a structural advantage in AI scaling.
  

A new set of industry reports shows that Chinese open-source large language models (LLMs) are reshaping the global AI landscape. Once considered secondary players, Chinese models have now reached almost 30% of worldwide usage and are gaining traction among developers, startups, and major companies in the U.S.

Explosive Growth in Global Adoption

Chinese open-source models grew from just 1.2% of global usage in late 2024 to nearly 30% in 2025, according to a study covering 100 trillion tokens. While Western proprietary systems such as OpenAI’s GPT-4o and GPT-5 still dominate with about 70% market share, Chinese models now match the usage of all other non-Chinese open-source models combined.

Key drivers include Alibaba’s Qwen series, DeepSeek V3, and Moonshot AI’s Kimi K2, all of which have seen rapid update cycles and widespread adoption among developers worldwide.

Strong Endorsements From Silicon Valley

Several high-profile U.S. technology leaders are publicly shifting toward Chinese LLMs. Airbnb CEO Brian Chesky said the company prefers Alibaba’s Qwen because it is “fast and cheap.” Investor Chamath Palihapitiya also moved key workflows to Moonshot’s Kimi K2, calling it more performant than alternatives.

On platforms such as Hugging Face and OpenRouter, Chinese models now dominate trending charts and usage rankings, replacing many U.S. systems.

Why Startups Prefer Chinese AI

Cost is a major factor. Chinese models often operate at one-fifth the price of Western models. MiniMax’s M2, for instance, offers performance similar to Claude Sonnet 4.5 at just 8% of the cost.

Open-weight licensing is another benefit, allowing companies to access model weights, fine-tune systems, and build custom versions more easily than with closed U.S. models. Alibaba reports more than 170,000 derivative models built on Qwen alone.

Geopolitical Concerns and Regulatory Tensions

The surge of Chinese AI tools has sparked debate in Washington. U.S. authorities warn of potential security risks associated with foreign AI systems, placing firms such as Zhipu AI on trade blacklists. Analysts say the rise of Chinese open-source AI could challenge long-standing U.S. dominance if current trends continue.

Despite concerns, a recent global survey shows 82% of users would adopt Chinese models if hosted outside China, indicating broad openness to using these systems.

China’s Infrastructure Advantage Raises Alarms

Nvidia CEO Jensen Huang recently warned that China’s ability to build AI infrastructure “at unbelievable speed” is a strategic advantage. He noted that China has double the energy capacity of the U.S., with much faster growth.

Massive nationwide data-center construction, supported by low-cost energy and government initiatives, is accelerating the country’s AI expansion.

Rapidly Shifting AI Balance

With competitive performance, lower prices, and fast iteration, Chinese models are becoming central to global AI development. Industry experts say the momentum marks a shift toward a two-superpower AI race, with China emerging as a formidable peer to the United States.

Read More:

• Say Goodbye to Driving: Rivian CEO Says Cars Will Soon Run Errands Without You
• ByteDance’s Agentic AI Phone Sparks Panic and Nationwide App Bans
• Leaked Video Exposes Tesla’s Robot Secret: Was Optimus Being Remote-Controlled?

(via)

The post Why U.S. Startups Are Dumping Western AI for China’s Open-Source Models appeared first on Gizmochina.

Just a year after surprising the industry with a model that shook up global markets, China’s DeepSeek is back with two new releases — and some ambitious comparisons. The latest open-source models, DeepSeek V3.2 and V3.2-Speciale, can reportedly match or even outperform the most powerful AI systems currently available, including OpenAI’s GPT-5 and Google’s Gemini 3 Pro.

Rather than chasing scale at any cost, DeepSeek continues to push a strategy based on efficiency. While American labs rely on massive clusters of cutting-edge chips, DeepSeek argues its refined training approach can deliver similar intelligence on more accessible hardware. According to the company, even the standard model includes native tool-use reasoning — meaning users get structured thinking capabilities even without switching into a dedicated reasoning mode.

The spotlight, however, is on V3.2-Speciale. DeepSeek claims this version has surpassed GPT-5 on internal benchmarks and sits on par with Gemini 3 Pro in reasoning-heavy tasks. As evidence, the company points to strong showings in the 2025 International Mathematical Olympiad and International Olympiad in Informatics, saying its final entries are published for anyone to inspect.

DeepSeek credits its jump in performance to two main innovations: a custom sparse-attention mechanism designed for long-context efficiency, and an expanded reinforcement learning pipeline trained on more than 85,000 complex, multi-step tasks — all created through its in-house “agentic task synthesis” system.

DeepSeek V3.2 is already available

For those eager to try the models, V3.2 is already available through DeepSeek’s website, mobile apps, and API. The more experimental V3.2-Speciale is accessible only via a temporary API endpoint that is set to be removed after December 15, 2025. It currently runs as a reasoning-only engine, without tool calling.

While independent benchmarking will ultimately determine how these systems stack up against the giants, one thing is increasingly clear: DeepSeek is determined to prove that top-tier AI doesn’t have to come with a top-tier price tag. And that message alone is putting pressure on the rest of the industry to rethink what’s possible — and what’s necessary — to stay competitive.

Don’t miss a thing! Join our Telegram community for instant updates and grab our free daily newsletter for the best tech stories!

For more daily updates, please visit our News Section.

(Source)

The post DeepSeek strikes again: V3.2 and V3.2-Speciale claim to challenge leading AI systems appeared first on Gizmochina.

Chinese startup DeepSeek is preparing to launch a next-generation AI agent by the end of 2025, according to a Bloomberg report. The Hangzhou-based company is developing a system that can perform complex, multi-step tasks with minimal input from users. The model will also learn from its past actions to improve performance over time.

DeepSeek gained global attention earlier this year with the release of its R1 model. The system demonstrated advanced reasoning capabilities and was built at a reported cost of just $6 million, far less than rivals like OpenAI’s ChatGPT or Google’s Gemini. R1 was also open source, allowing developers to freely access its code. The release disrupted the “bigger is better” approach in Silicon Valley, where massive budgets and vast datasets have dominated AI development.

Since R1, DeepSeek has moved cautiously. Founder Liang Wenfeng delayed the R2 model, a move attributed to technical refinements and his ongoing work at High-Flyer Asset Management. This deliberate pace differs from competitors like Alibaba and Tencent, which have accelerated their own AI rollouts.

Last month, DeepSeek rolled out its V3.1 update, expanding the context window to 128,000 tokens and increasing the parameter count to 685 billion. The company also now applies mandatory AI-generated content labels on all outputs, a policy that cannot be disabled and sets DeepSeek apart from its peers.

The upcoming R2 model will target the emerging market for AI agents. Unlike chatbots that deliver text-based responses, agents are designed to handle tasks such as travel planning, software debugging, and business workflows with little human supervision. Analysts view this technology as the next major step in AI, capable of boosting productivity and reshaping digital services.

Global competitors, including OpenAI, Microsoft, and Anthropic, have already introduced agent-focused features. DeepSeek aims to match and potentially surpass those efforts, continuing its push to challenge U.S. dominance in AI.

Details about pricing, availability, and technical specifications remain limited. However, industry watchers expect the launch to be closely monitored in both Silicon Valley and Washington, where DeepSeek’s rapid progress has already raised concerns about China’s growing influence in artificial intelligence.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

(Source)

The post DeepSeek is working on AI agent to take on OpenAI by end of 2025 appeared first on Gizmochina.

DeepSeek, the AI company, has just rolled out new rules in China requiring all AI-generated content on their platform to be clearly marked as artificial. This change comes as part of broader efforts to make AI content more transparent and prevent people from misusing it.

The company sent out an official notice explaining that anything created by their AI systems must now include clear signs showing it wasn’t made by humans. These markers come in two forms: obvious ones that people can easily see, and hidden technical ones built into the content itself.

The visible markers might be text saying “AI-generated,” audio announcements, or graphics that appear somewhere in the content. The hidden markers are tucked away in the technical data and include details like what type of content it is, which company made it, and a unique ID number.

Users aren’t allowed to mess with these labels in any way. They can’t remove them, change them, fake them, or try to hide them. DeepSeek has also banned any tools or services that would help people tamper with the labels. If someone breaks these rules, they could face legal consequences.

Along with this announcement, DeepSeek also released a detailed technical guide explaining how their AI actually works. The document covers things like how they train their models, what data they use, and the process behind creating content. They say this is meant to help people understand the technology better, use it more responsibly, and give everyone the right to know where content comes from.

These new requirements aren’t just DeepSeek’s idea. They are following the new Chinese government regulations. The rules were put in place by several government agencies and require all AI-generated content to have labels that can be tracked and cannot be tampered with. The companies providing these AI services are responsible for making sure everyone follows the rules.

In related news, DeepSeek has recently released V3.1 with a 128K token context and 685B parameters, unifying reasoning and general tasks while the R2 model remains delayed. Meanwhile, Nvidia’s GB300 GPUs reportedly achieve 6x the performance of H100s on DeepSeek R1, offering major efficiency and scalability gains for enterprise AI.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

(Via)

The post DeepSeek now slaps AI labels on all content and you can’t remove them appeared first on Gizmochina.

NVIDIA’s latest Blackwell-based GB300 GPUs are starting to show what they can do, and early results point to a massive jump in efficiency compared to the company’s previous generation. A recent CoreWeave benchmark shows just how far the new chips push AI performance when running one of today’s heaviest workloads.

In tests using the DeepSeek R1 model, just four GB300 GPUs matched the performance of 16 H100s. That works out to roughly six times the throughput per card, thanks to both raw hardware upgrades and a more efficient architecture.

The GB300 NVL72 platform can scale up to 37TB of memory (40TB max) with 130TB/s of memory bandwidth, allowing it to tackle extremely large AI models without the bottlenecks seen in older hardware. It also makes use of 4-way tensor parallelism (TP4), while the H100 required 16-way splits (TP16) to reach similar scale. Fewer splits mean less communication overhead, and NVIDIA’s fifth-gen NVLink and NVSwitch interconnects help keep latency low.

For businesses running large AI services, that means faster token generation, lower costs per inference, and a more straightforward path to scaling. CoreWeave noted that the efficiency gains translate directly into practical benefits for workloads such as reasoning models, where performance-per-watt and latency are critical.

Of course, the hardware won’t come cheap. Systems built on the GB300 NVL72 are expected to start at around $300,000, putting them firmly in enterprise territory. But for companies already investing heavily in AI infrastructure, the performance leap could justify the cost.

If these early numbers hold true as deployments expand, the GB300 could mark one of the biggest generational jumps NVIDIA has delivered yet — and a new baseline for high-end AI computing.

Don’t miss a thing! Join our Telegram community for instant updates and grab our free daily newsletter for the best tech stories!

For more daily updates, please visit our News Section.

The post NVIDIA GB300 GPUs deliver huge AI efficiency gains in Deepseek R1 inference test appeared first on Gizmochina.

Tesla has introduced a new AI-powered voice assistant in its electric vehicles in China, developed in partnership with DeepSeek and ByteDance. The feature is called “Hey Tesla” and is designed to give drivers natural voice control for navigation, media, and cabin functions.

Tesla confirmed that ByteDance’s Doubao large language model will manage voice commands such as adjusting the air conditioning, setting routes, and controlling entertainment. DeepSeek’s chatbot will handle conversational tasks, including news, weather updates, and casual interaction. Both models operate on ByteDance’s Volcano Engine cloud platform, which processes requests in real time through an encrypted API.

The company launched the system with the new six-seat Model Y L SUV, which went on sale in August. Unlike older Tesla models in China that require pressing a steering wheel button, the Model Y L supports a wake phrase. Drivers can say “Hey Tesla” or set a custom word to activate the assistant hands-free.

Tesla’s choice of partners reflects its push to localize features for the Chinese market. Domestic rivals such as BYD, Nio, and Xpeng already use advanced AI voice assistants, giving them an advantage in customer experience. By adopting DeepSeek and Doubao, Tesla aims to close that gap and compete more directly with homegrown brands.

In the United States, Tesla continues to use Grok, the AI model built by Elon Musk’s startup xAI. The split shows Tesla’s regional strategy, with local AI providers powering its cars in China to meet regulatory and consumer expectations. BMW has taken a similar path, working with Alibaba’s QWen LLM for its Chinese lineup.

Tesla has not confirmed when the new voice assistant will expand to other models in China. The company’s updated terms of service mention the integration, but some owners said they had not received the feature in recent over-the-air updates. Despite the delay, Tesla is signaling that in-car intelligence is now central to competing in the world’s largest EV market.

In related news, we recently covered a comprehensive guide on AI chatbot subscriptions in 2025, including all the key plans, pricing, and features you need to know.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

(Via)

The post Tesla partners with DeepSeek and ByteDance to launch AI voice assistant in China appeared first on Gizmochina.

AI chatbots have become subscription services much like music and video streaming platforms. Each company now has different pricing tiers designed to attract casual users, professionals, and enterprises. Here is a detailed look at the major players and what they offer.

ChatGPT (by OpenAI)

OpenAI continues to lead the market with ChatGPT. The free plan gives users access to GPT-5 with usage limits, enough for simple queries and casual use.

The ChatGPT Plus plan costs $20 per month globally or ₹1,999 in India. It unlocks unlimited use of GPT-5 and GPT-4o, advanced tools such as web browsing, data analysis, and image generation, as well as custom GPT creation. Plus subscribers also receive early access to new features.

The ChatGPT Pro tier, priced at about $200 per month or ₹19,900 in India, is aimed at heavier workloads. It increases usage capacity, adds customization, and provides enterprise-grade security.

OpenAI also introduced ChatGPT Go in select regions like India at ₹399 per month. It delivers GPT-5 access with higher limits than the free tier but does not include advanced features such as Sora video generation.

At the enterprise level, ChatGPT Enterprise offers unlimited GPT-5 and GPT-4o access with no usage caps, extended context windows, SOC 2-compliant security, and admin tools. Pricing is custom-based on business needs.

Gemini (by Google)

Google’s Gemini is closely tied to its ecosystem. The free tier allows access to Gemini 2.5 Flash and Gemini 2.5 Pro for basic conversations.

The main paid plan is Gemini Advanced, part of the Google One AI Premium subscription priced at $19.99 per month. This plan includes 2TB of cloud storage and unlocks Gemini 2.5 Pro with higher limits, a 1 million–token context window, file analysis features, and video generation powered by Veo 3.

Gemini Ultra is available at $249.99 per month for users who want the most powerful access, including experimental features and video generation with sound.

For enterprises, Google provides Gemini access through Google Cloud with custom pricing and token-based API billing.

Microsoft Copilot

Microsoft integrates Copilot across Windows, Office, and mobile apps. The free version provides basic assistance.

Copilot Pro, priced at $20 per month, adds priority access to OpenAI’s latest models, faster image creation with Microsoft Designer, and direct integration with Word, Excel, and PowerPoint. A Microsoft 365 subscription is required for Office features.

Copilot for Microsoft 365 costs $30 per user per month and extends Copilot into the full Microsoft suite with enterprise-grade security.

Perplexity

Perplexity positions itself as an AI-powered search engine. The free tier gives limited Pro Searches each day.

Perplexity Pro costs $20 per month or $200 annually. Subscribers receive over 300 Pro Searches daily, access to GPT-5 and Claude, unlimited file uploads, and image generation.

The Enterprise Pro plan costs $40 per user per month and adds private knowledge hubs, administrative controls, and strict privacy protections.

Claude (by Anthropic)

Claude is popular for long-form writing and research tasks. The free plan offers basic access to the AI.

Claude Pro costs $20 per month and expands usage limits, giving access to Claude Sonnet 4 and extended conversations.

Claude Max starts at $100 per month and provides much higher usage, larger output capacity, and early access to new features.

API pricing ranges from $0.25 per 1M tokens for Claude Haiku to $75 per 1M output tokens for Claude Opus.

Grok (by xAI)

Grok, developed by Elon Musk’s xAI, is integrated into the X platform. There is no free version. Access is bundled with X Premium+ at $16 per month.

The chatbot delivers real-time answers based on X’s live feed and uses a witty, sarcastic style. It is less about productivity and more about social interaction.

Jasper

Jasper is built for content marketing rather than general conversation. There is no free plan.

The Creator plan costs $49 per month, while the Pro plan is priced at $69 per month. Enterprise users can negotiate custom pricing under the Business plan.

Jasper includes more than 50 templates for content creation, SEO optimization, and integrations with marketing platforms.

Poe (by Quora)

Poe acts as a central hub for AI chatbots from multiple providers. The free plan provides limited access to bots like GPT-5, Claude Opus 4.1, and DeepSeek-R1.

Poe Pro, at $19.99 per month, removes most restrictions and allows unlimited use of premium models such as GPT-5 and Claude Sonnet.

DeepSeek

DeepSeek has gained recognition for its reasoning and coding performance. The chatbot is free for consumers.

Developers can use its API with extremely low pricing. The deepseek-chat model costs $0.27 per 1M input tokens and $1.10 per 1M output tokens, making it one of the most affordable large-scale AI solutions.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

The post AI Chatbot Subscriptions in 2025: Everything You Need to Know appeared first on Gizmochina.

Chinese AI company DeepSeek has released version 3.1 of its flagship large language model, expanding the context window to 128,000 tokens and increasing the parameter count to 685 billion. The update was announced quietly through the company’s WeChat user group on August 19, without any posts on its official social media channels.

What’s New?

The most significant change in DeepSeek V3.1 is the increased context length, which now allows the model to handle inputs equivalent to a 300 to 400-page book. This improvement enables better performance in long-form content generation, technical document analysis, and extended multi-turn conversations. The official group confirmed the expanded context was already supported internally in the previous V3 version, but has now been formally enabled across all interfaces.

Mixture-of-Experts and Benchmark Scores

DeepSeek V3.1 continues to use a Mixture-of-Experts (MoE) architecture, with only 37 billion parameters activated per token. The model supports multiple precision formats, including BF16, FP8, and F32, providing more flexibility for different deployment environments. Developers can access the model via API or download it from Hugging Face under the MIT open-source license.

The upgraded model performed well in early third-party benchmarks. It scored 71.6% on the Aider coding test, placing it above Claude Opus 4 and making it one of the strongest open-source coding models currently available. DeepSeek V3.1 also showed improved performance in math and logic tasks. However, some users noted no clear gains in reasoning compared to the earlier R1-0528 model.

Shift in Strategy

DeepSeek has removed all references to the R1 model from its chatbot interface, signaling a shift toward a single hybrid model architecture. The company appears to have integrated its reasoning capabilities into V3.1, instead of maintaining a separate reasoning model.

The training cost for V3.1 has not been disclosed. However, according to previous reports, the original V3 model was trained on 2.788 million GPU hours using Nvidia H800 chips, at an estimated cost of $5.6 million. That model formed the base for the current version, which likely shares similar infrastructure with additional refinements.

Confusion Around the Delayed R2 Model

There had been widespread anticipation that DeepSeek’s next major release would be the long-awaited R2 model, designed to advance reasoning capabilities. Instead, V3.1 appeared as the company’s next step. According to a recent Financial Times report, the R2 model’s release has been delayed due to persistent technical issues involving Huawei’s Ascend AI chips.

DeepSeek was reportedly urged to use Ascend hardware to reduce reliance on Nvidia, aligning with China’s national strategy for AI self-sufficiency. Despite support from Huawei engineers, training on Ascend failed due to compatibility and performance issues. The company then switched to using Nvidia GPUs for training while retaining Ascend for inference. This hybrid setup introduced further complications and delays. In addition, the extended time spent on data labeling slowed down development. DeepSeek founder Liang Wenfeng reportedly expressed frustration with the slow progress.

Meanwhile, competitors such as Alibaba’s Qwen3 have moved ahead by deploying similar algorithms with more efficient execution. The episode has underscored the limitations of China’s domestic chip infrastructure and the challenges faced by startups attempting to meet political and technical demands simultaneously.

DeepSeek has not ruled out the launch of R2. However, whenever the model does arrive, its performance will face intense scrutiny. Until then, V3.1 stands as the company’s current flagship, serving both reasoning and non-reasoning workloads in a unified framework.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

The post DeepSeek Releases V3.1 Model: What’s New? appeared first on Gizmochina.

Earlier today, we reported that DeepSeek’s next-generation large language model, R2, was rumored to launch later this month, based on speculation circulating in Chinese tech circles and even hinted at by DeepSeek’s own AI.

The model, in response to user queries, suggested a release window between August 15 and 30. But just hours after this surfaced, a source close to DeepSeek has categorically denied the claim, confirming that there are no plans to launch DeepSeek-R2 in August.

What’s Slowing Down DeepSeek-R2?

This isn’t the first time DeepSeek’s roadmap has stirred confusion. Similar rumors made the rounds earlier this year with an alleged March 17 release, which also turned out to be false. Despite the public anticipation, the company has yet to officially confirm a release date or even disclose key technical specifications of the R2 model.

That hasn’t stopped speculation. DeepSeek-R2 is widely expected to be a major leap forward, improving multilingual reasoning and code generation, and rivaling top-tier models like GPT-4/5. But progress has reportedly hit a wall. According to insiders cited by The Information in June, DeepSeek CEO Liang Wenfeng remains dissatisfied with the model’s performance, even after months of intense development.

The delay isn’t just a matter of refinement; it may also reflect infrastructure concerns. DeepSeek has been operating under limited compute availability, especially after Nvidia’s H20 chips were banned from export to China. Sources suggest the company fears launching R2 too soon could backfire if usage spikes overwhelm its systems. It’s a real concern in the AI race, where performance alone isn’t enough; scalability matters too.

To address the shortfall, DeepSeek is believed to be training R2 on Huawei’s Ascend 910B chips. Huawei’s compute cluster reportedly delivers 512 PFLOPS of FP16 performance at 82% utilization, equivalent to about 91% of the performance of Nvidia’s A100 cluster. While this shows promise, it’s unclear whether the setup can meet the scale required for a major model rollout.

Meanwhile, the chip supply picture may be improving. Just days ago, Nvidia and AMD agreed to a revenue-sharing deal with the US government, allowing them to continue selling AI chips in China in exchange for handing over 15% of their China revenue. The agreement may eventually ease supply constraints for companies like DeepSeek. However, any immediate benefit remains uncertain, as Chinese authorities have reportedly discouraged domestic firms from using Nvidia’s H20 chips.

So while DeepSeek’s own model might have teased an imminent release, the company’s actual position appears far more cautious. R2 is real. It’s under active testing. But it’s not ready for prime time. And in a hyper-competitive AI landscape, DeepSeek needs more than a good model; it needs one that’s stable, scalable, and market-ready.

Until then, the spotlight stays on DeepSeek, as the world waits to see if R2 can live up to the growing expectations.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

(Via)

The post DeepSeek-R2 August launch hype dismissed by insider as pressure mounts on AI ambitions appeared first on Gizmochina.

DeepSeek is preparing to launch its next-generation large language model, DeepSeek-R2, sometime between August 15 and August 30, 2025. The launch window came to light during an interaction with DeepSeek’s own AI model, hinting at an imminent rollout just days after the release of ChatGPT-5.

DeepSeek-R2 is expected to make a significant leap in architecture by adopting a more advanced Mixture of Experts (MoE) setup. It will also integrate a smarter gating network to better handle inference-heavy workloads. According to sources familiar with its development, the model could scale up to 1.2 trillion parameters, nearly double that of DeepSeek-R1, which had 671 billion. This will still be lower than ChatGPT-4/5, which has over 1.8 trillion parameters.

China looking for domestic AI self-sufficiency

In a strategic move toward domestic AI self-sufficiency, the model has been entirely trained on Huawei’s Ascend 910B chips. Huawei’s compute cluster, delivering 512 PFLOPS of FP16 performance at 82% utilization, reportedly achieves 91% of the performance of Nvidia’s A100 cluster. Analysts see this as a critical step in reducing China’s reliance on US-made AI hardware.

DeepSeek-R2’s training reportedly cost 97% less than GPT-4, thanks to local hardware and optimization techniques. Analysts expect DeepSeek to offer API access at lower prices, which could shake up current pricing models dominated by OpenAI and Anthropic.

The anticipation around DeepSeek-R2 has already sparked movement in Chinese tech stocks. AI chipmaker Cambricon saw a 20% surge in its share price, pushing its market cap past 355 billion yuan (around $49.7 billion).

In related news, Huawei has rolled out a new AI inference framework called Unified Cache Manager (UCM). Designed to accelerate model inference, UCM optimizes how KV Cache data is handled across memory tiers like HBM, DRAM, and SSDs. In tests with China UnionPay, Huawei reported up to a 90% reduction in latency and a 22x boost in throughput. The company plans to open-source UCM in September.

Together, the launch of DeepSeek-R2 and the introduction of Huawei’s UCM framework signal a major shift in China’s AI ambitions. These developments point to a future where China builds and runs high-performance AI systems without relying on Western chips or software tools.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

(Via)

The post DeepSeek-R2 may launch late August as China pushes for AI independence appeared first on Gizmochina.

South Korea has temporarily halted new downloads of the Chinese AI chatbot DeepSeek, citing concerns over data privacy violations. The decision, announced by the Personal Information Protection Commission (PIPC), took effect on February 15, 2025, at 6:00 p.m. local time. While the mobile app downloads have been suspended, the web version remains accessible as the company works on meeting regulatory requirements.

South Korea Cracks Down on DeepSeek AI

The PIPC revealed that its analysis of DeepSeek, conducted soon after the chatbot’s launch, uncovered flaws in its communication features and policies regarding the handling of personal information with third-party service providers. The commission emphasized that until these issues are resolved in compliance with South Korea’s Personal Information Protection Act, the app will remain unavailable for new downloads. Existing users have been advised to refrain from inputting personal data into the chatbot’s prompts until further notice.

DeepSeek has acknowledged that it failed to fully consider South Korean data protection laws before launching and has since appointed a local representative to facilitate regulatory compliance. The government aims to use this case as a precedent to strengthen guidance and oversight, ensuring that similar lapses do not occur in the future.

The suspension follows heightened scrutiny from South Korea’s National Intelligence Service (NIS), which previously warned that DeepSeek was excessively collecting user data and potentially using it to train its AI models. Additionally, security experts discovered that the app’s Android and iOS versions transmitted certain user data to its servers in an unencrypted format, raising further concerns.

This is not an isolated case, as global apprehension over DeepSeek’s data practices has been growing. The United States, Italy, Australia, and Taiwan have already imposed restrictions on the AI service for government use, citing security risks. NASA blocked DeepSeek from its systems, and the U.S. Navy warned personnel against using the app due to potential data vulnerabilities.

Amid these mounting concerns, Beijing maintains that it allows international internet companies to operate in China under local laws and insists that it does not compel firms to collect or store data unlawfully. However, as regulatory scrutiny intensifies worldwide, DeepSeek now faces an uphill battle to regain trust and compliance across multiple jurisdictions.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

(Via)

The post South Korea suspends DeepSeek AI downloads amid privacy backlash appeared first on Gizmochina.

Nubia has unveiled the world’s first smartphone to feature full-scale DeepSeek AI integration, setting a new milestone in on-device artificial intelligence. According to the company, their current flagship Nubia Z70 Ultra incorporates the DeepSeek model at a system-wide level, eliminating the need for standalone apps while enabling fluid AI-driven interactions.

Unlike traditional AI assistants that rely on cloud processing or require dedicated applications, DeepSeek’s integration in the Z70 Ultra allows users to access its capabilities instantly.

This includes seamless voice assistant functionality and Nebula AI-powered image recognition, which lets users snap a photo to analyze and solve complex problems in real-time. Nubia has also extended its AI ecosystem beyond smartphones, incorporating real-time activation features in its earbuds to enable continuous conversation support. In addition to the Z70 Ultra, other models will soon receive updates.

Beta Update Rolls Out

Ahead of today’s announcement, Nubia had already begun rolling out a beta update to Z70 Ultra users. The 126MB update enhances DeepSeek integration across both the standard and Starry Sky editions, refining voice assistant capabilities and optimizing Future Mode, which now defaults to enabling Nebula AI.

Additional refinements include adjustments to the search bar’s position, fixes for app search redirections, and improvements to the screen assistant’s UI behavior. The update also resolves memory leaks in the Nebula Gravity system, addressing stability concerns.

DeepSeek Gains Momentum

Nubia isn’t the only brand looking to embed DeepSeek at the core of its devices. Other Chinese manufacturers are reportedly developing similar integrations, while Samsung is rumored to be exploring DeepSeek AI for its Galaxy smartphones in China.

According to leaker Ice Universe, this move could position DeepSeek as a regional alternative to Google’s Gemini AI, enhancing One UI with localized AI-driven features. Beyond functionality, Samsung’s adoption could also help the company navigate China’s regulatory landscape and compete more effectively with AI-heavy rivals like Huawei. While Samsung has yet to confirm its plans, DeepSeek’s growing adoption suggests it could soon become a standard feature in Chinese-market smartphones.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

The post Nubia Z70 Ultra becomes the first smartphone to feature full-scale DeepSeek AI appeared first on Gizmochina.

DeepSeek is a Chinese AI company that has shaken up the industry with its efficiency claims. The company claims training their V3 model, the predecessor of the R1 model that everyone is using, costs just $5.576 million to train.

Since the release of the R1 model, DeepSeek has achieved widespread adoption all over the world and now Arm CEO predicts that the US government could soon ban the Chinese AI company. National security? Let’s explore what could be the reason.

According to a previous report, NASA has already blocked DeepSeek from its systems, and the U.S. Navy has warned personnel against using the AI service.

DeepSeek is reportedly 50-75% more efficient than its competitors. Despite requiring less power and compute, it has outperformed the OpenAI’s o1 model in different tests. Additionally, it’s free and open source, meaning anyone can build their AI application on top of it.

Privacy? One can download the model and run it locally on their own system. The 7b or 8b parameter model (both are less than 5GB in size) can be installed and operated on a consumer grade device.

A possible reason behind the prediction

DeepSeek has achieved a widespread adoption throughout the world (including the US) and triggered a major selloff in AI-related stocks, with Nvidia losing $600 billion in market value over fears that lower-cost AI models could reduce the demand for expensive AI chips.

It not only achieved the top position in Apple’s app store but also convinced US-based giants like Microsoft to adopt it—companies that would have otherwise stuck with OpenAI for its AI needs. Perplexity has also integrated DeepSeek R1 for better reasoning capabilities and overall smarter responses, which they are running on their servers.

We will need to see if the prediction turns out to be true and how the US companies that are already using or working on it navigate the situation.

Not to mention a lot of the Chinese smartphone makers such as Huawei, Honor, Oppo, and ZTE have also introduced the collaboration with DeepSeek and are working on distilled versions of the R1 model for their on-device AI needs.

Tech enthusiast? Get the latest news first! Follow our Telegram channel and subscribe to our free newsletter for your daily tech fix!

For more daily updates, please visit our News Section.

(Source | Via)

The post DeepSeek could soon face a ban in the US, predicts ARM CEO appeared first on Gizmochina.

Samsung launched its latest flagship with the entire focus on new or improved AI features. Now the Chinese smartphone makers are also trying to catch up or surpass Samsung in the AI race, thanks to China’s DeepSeek. Several OEMs including Huawei, Honor, and Oppo have already been reported collaborating with DeepSeek to make their AI assistants smarter and more capable. Now Nubia has also announced their collaboration with DeepSeek.

For those unfamiliar, DeepSeek not only showcases China’s progress in the field of AI, but it’s the biggest shake-up in the field since OpenAI’s launch of ChatGPT. DeepSeek is more efficient to operate (this is significant because big tech giants were already building nuclear power plants to power their AI servers) outperforms OpenAI’s GPT-4o in benchmarks and is open source, meaning anyone can use it.

Nubia to integrate DeepSeek in its Nebula AIOS

According to Nubia President Ni Fei (via ITHome), the company is aiming for a deeper integration of DeepSeek within its UI skin. The company has been increasingly focusing on AI integration for its phones and this collaboration might be a big step in the right direction. The official suggests that unlike other OEMs collaborating with DeepSeek, Nubia is going to “deeply embed DeepSeek into the system” which will reportedly utilize the full-scale 671 billion parameters.

It seems like the company is looking to run the model on-device, which has a number of benefits such as no internet requirement and data privacy.

The latest flagship, the Nubia Z70 Ultra, is going to be the first to benefit from the DeepSeek integration, and the development is already in progress. This phone packs the Snapdragon 8 Elite chip, a 6150mAh Si/C battery, and supports 80W charging.

Tech enthusiast? Get the latest news first! Follow our Telegram channel and subscribe to our free newsletter for your daily tech fix!

For more daily updates, please visit our News Section.

(Source)

The post Nubia Z70 Ultra could run DeepSeek on-device appeared first on Gizmochina.

Several countries have taken action against DeepSeek, a Chinese artificial intelligence (AI) platform, over concerns related to national security, data privacy, and possible links to the Chinese government. Authorities fear that DeepSeek could pose a security risk due to its origins and the potential for data transfers to Chinese state entities.

National Security and Data Privacy Risks

DeepSeek has been banned in multiple countries primarily due to security and privacy concerns. Governments worry that the Chinese government could access user data collected by DeepSeek under China’s National Intelligence Law. There are concerns over the platform’s lack of transparency in handling user data, raising fears of possible data breaches or unauthorized data sharing. 

Which countries have banned DeepSeek

Australia
Australia has banned DeepSeek on all government devices due to national security risks. The Department of Home Affairs ordered all government agencies to remove DeepSeek, with Home Affairs Minister Tony Burke emphasizing that the ban aims to protect Australia’s national interests. 

South Korea
South Korea has prohibited its use on government employees’ devices across multiple ministries and state-run agencies due to security and privacy concerns. Several government agencies, including Korea Hydro & Nuclear Power, have blocked AI services like DeepSeek, while the country’s Personal Information Protection Commission is investigating how DeepSeek manages user data.

Italy
Italy has also taken steps to restrict DeepSeek, particularly regarding the processing of Italian users’ data. The Italian Data Protection Authority (Garante) imposed restrictions and requested details on DeepSeek’s data policies, citing a lack of transparency in data handling. 

Taiwan
Taiwan has banned DeepSeek from all government departments, citing national security risks, and has blocked it from all government networks.

United States is considering a ban

The United States has not yet banned DeepSeek but is considering restricting its use. NASA has already blocked DeepSeek from its systems, and the U.S. Navy has warned personnel against using the AI service. Additionally, U.S. lawmakers are drafting a bill to ban DeepSeek on government-owned devices due to security concerns.

What Data does DeepSeek collect?

According to its privacy policy, DeepSeek collects personal information such as email, phone number, password, and date of birth. It also records chat history, including user interactions with the chatbot. Technical data such as IP addresses, keystroke patterns, and operating system details are also gathered. Information is shared with service providers and advertising partners, raising concerns over how long data is retained and who has access to it.

How DeepSeek data policy compares to other AI platforms

DeepSeek’s data collection practices are not unique. Other AI platforms like OpenAI’s ChatGPT also gather user data. However, OpenAI follows stricter data protection regulations in regions like the EU. Western governments believe OpenAI can be regulated, whereas Chinese platforms like DeepSeek are perceived as direct security threats.

What is the future of DeepSeek?

DeepSeek has disrupted the AI market with its cost-effective model. While OpenAI’s GPT-4 required over $100 million in training costs, DeepSeek’s latest model was developed for less than $6 million. Its rapid rise in popularity even caused a 17% drop in Nvidia’s stock value, leading to significant shifts in global AI investments.

As more governments evaluate DeepSeek’s risks, additional bans could be implemented. Western governments continue to assess the security implications of Chinese AI platforms, emphasizing concerns about espionage and unauthorized data access. The controversy surrounding DeepSeek highlights the broader tension between AI development and national security policies worldwide.

Read More:

• All you need to know about DeepSeek: China’s AI game changer and a wake-up call for US
• DeepSeek may be the answer to Apple’s AI woes in China: Apple Intelligence for iPhone finally found a way in?
• Huawei and Alibaba cloud integrate DeepSeek AI models
• India working on affordable AI models to rival ChatGPT & DeepSeek

(via)

The post Countries are banning DeepSeek: Is the Chinese AI a threat ? appeared first on Gizmochina.

Huawei Cloud has partnered with AI startup SiliconFlow to bring DeepSeek’s AI models to its Ascend cloud service. This partnership makes AI more affordable and accessible. The two main models, DeepSeek V3 (a language model) and DeepSeek R1 (a reasoning model), offer high performance at a lower cost.

Affordable AI Access with DeepSeek

Huawei Cloud offers discounted rates for DeepSeek AI models through SiliconFlow. The pricing is competitive, with the DeepSeek V3 model costing only 1 yuan ($0.13) per million input tokens and 2 yuan ($0.26) per million output tokens. The DeepSeek R1 model is priced at 4 yuan ($0.53) per million input tokens and 16 yuan ($2.13) per million output tokens. These affordable rates make AI access more widely available.

AI Independence and Geopolitical Impact

Huawei’s deployment of DeepSeek models reflects China’s push for AI self-sufficiency amid U.S. restrictions on high-end chips. The models run on Huawei’s Ascend AI hardware, reducing dependency on foreign technology. This move strengthens China’s AI ecosystem and positions it as a strong competitor in global AI development.

Alibaba joins the DeepSeek AI wave

Following Huawei’s integration, Alibaba Cloud has also launched DeepSeek AI services on its platform. Users can access these models through the PAI Model Gallery, allowing them to deploy AI without coding. Alibaba Cloud offers both full and distilled versions of the DeepSeek R1 model, providing flexibility for different AI applications.

DeepSeek’s growth and future prospects

DeepSeek AI is gaining popularity in China and the U.S., with Microsoft and Amazon integrating its models into their platforms. While DeepSeek trained its AI models on Nvidia’s H100 GPUs, it now runs inference on Huawei’s Ascend 910C chip. Huawei is also developing the Ascend 920C, expected to match Nvidia’s upcoming Blackwell B200, further improving DeepSeek’s AI capabilities.

DeepSeek’s rapid expansion and integration with major cloud platforms signal China’s growing influence in AI. With cost-effective models and improved hardware, DeepSeek is emerging as a major competitor in the global AI race.

Read more:

• All you need to know about DeepSeek: China’s AI game changer and a wake-up call for US
• China’s DeepSeek AI Assistant is now the top free app for iPhones
• Amap unveils China’s first English map: A game-changer for foreign tourists
• China’s first AI railway robot is here and it’s a game-changer

The post Huawei and Alibaba cloud integrate DeepSeek AI models appeared first on Gizmochina.

As ChatGPT’s dominance shows signs of decline, Chinese AI upstart DeepSeek is making a determined push into the global market. Recent Semrush data reveals that ChatGPT’s daily visits in the U.S. have dropped sharply from 22.1 million in October 2024 to 14.9 million by January 2025. Meanwhile, DeepSeek, though still far behind, has seen substantial growth, rising from just 2.3k visits in October to 71.2k by January 19, a significant surge in a short span.

Globally, DeepSeek’s traffic has surged past 7.12 million, and its rapid ascent to the top of download charts in 140 countries signals a notable shift in consumer preferences. However, its success has also sparked intense controversy.

DeepSeek’s Rapid Rise

At the heart of the debate is DeepSeek’s potential national security risks. Unlike OpenAI’s ChatGPT, DeepSeek operates under Chinese jurisdiction, subject to Beijing’s stringent data-sharing laws. Experts warn that the app’s vast trove of user data, ranging from personal queries to industry-specific intelligence could be exploited for strategic advantage. Cybersecurity concerns have already surfaced, with prominent firms identifying vulnerabilities and DeepSeek itself confirming a major cyberattack last week.

Dewardric McNeal, a senior analyst at Longview Global, underscores the broader implications: “This isn’t just about stolen data—it’s about mapping public sentiment, tracking industry trends, and influencing narratives.” The concern extends beyond individuals; industries and policymakers fear that DeepSeek’s open-source approach could allow China deeper insights into U.S. supply chains and technological developments.

Former Biden administration advisor Matt Pearl is even more alarmed, dismissing DeepSeek’s privacy policy as meaningless under Chinese law. He warns that the app could be leveraged for mass surveillance, tracking users across devices, and even embedding malware through updates. “One malicious update is all it takes,” Pearl cautions.

With tensions between the U.S. and China escalating, some believe a ban on DeepSeek is inevitable. “If TikTok faced scrutiny, DeepSeek is an even bigger concern,” Pearl argues. As AI competition heats up, the question remains: Will the U.S. allow a Chinese AI disruptor to dominate its market, or will regulatory action curb its rise?

You can also read our article on what makes DeepSeek different from other AI models.

For more daily updates, please visit our News Section.

Tech enthusiast? Get the latest news first! Follow our Telegram channel and subscribe to our free newsletter for your daily tech fix!

(Via)

The post DeepSeek gains momentum in the US as ChatGPT sees a dip appeared first on Gizmochina.

Artificial Intelligence has taken the world by storm with LLM (Large Language Models) being immensely popular for their diverse functionality. ChatGPT is a great example of an AI model, along with new disruptive models like DeepSeek from China. But now, it appears that India seeks to rival these with its own AI model, which could be arriving as early as this year.

Indian Govt to soon launch an affordable AI model

During a recent AI event, Ashwini Vaishnaw, the Union Minister of Electronics and Information Technology stated that India is working on its own foundational AI model. The minister further added that this will function similarly to DeepSeek and ChatGPT but for an affordable development cost. The government official stated that this new AI model could be ready in just 8 to 10 months.

In the event by the Indian AI Mission, Ashwini Vaishnaw revealed that researchers in India have been developing an AI ecosystem framework to support its own foundational AI model. This is being developed to offer an experience tailored to Indian users. It will also understand the linguistic and contextual needs of the Indian users, bringing inclusivity while eliminating biases.

The Union Minister of Electronics and Information Technology also talked about India’s computation prowess since the domestic AI model is being developed with a computational facility that employs 18,693 GPUs. To recall, ChatGPT was trained using around 25,000 GPUs, while DeepSeek was trained with 2,000 GPUs.

A typical popular AI model like ChatGPT costs about $3 to use for an hour, India’ AI model could cost just Rs 100 (roughly $1.15) thanks to government subsidy. This news also arrives after UC Berkeley researchers managed to replicate DeepSeek AI for only $30.

For more daily updates, please visit our News Section.

Tech enthusiast? Get the latest news first! Follow our Telegram channel and subscribe to our free newsletter for your daily tech fix!

The post India working on affordable AI models to rival ChatGPT & DeepSeek appeared first on Gizmochina.

AI research is usually a game for big tech companies with deep pockets. However, a team at UC Berkeley just flipped the script. They have replicated the core abilities of DeepSeek R1-Zero for just $30 (not a typo). Their project, called TinyZero, proves that advanced AI reasoning models don’t have to cost a fortune. Not to mention, AI research is becoming more accessible than ever.

Led by Jiayi Pan, the researchers aimed at recreating DeepSeek’s reasoning model using reinforcement learning (RL). Instead of relying on expensive cloud services or massive compute power, they trained TinyZero with a basic language model, a simple prompt, and a reward system.

Pan shared his excitement on X (formerly Twitter), saying, “You can experience the ‘Aha’ moment yourself for < $30.” He also described TinyZero as the first open reproduction of reasoning models, highlighting how it learned to verify and refine its own answers.

How TinyZero was developed

To test the model, the researchers used a game called Countdown, where players have to reach a target number using basic math operations. While at first, TinyZero guessed randomly, over time, it learned to verify its answers, search for better ones, and adjust accordingly.

They experimented with different model sizes, from 500 million parameters to 7 billion parameters. The results? Smaller models (0.5B parameters) just guessed answers and stopped. Larger models (1.5B+ parameters) learned to self-verify, refine solutions, and significantly improve accuracy.

What makes TinyZero truly impressive is how cheap it is compared to traditional AI models. Just look at these costs:

• OpenAI’s API: $15 per million tokens
• DeepSeek-R1: $0.55 per million tokens
• TinyZero’s total cost: $30—one-time training cost

This means anyone—not just big tech—can experiment with AI reasoning models without breaking the bank.

Availability

TinyZero is open-source and available on GitHub, so anyone can tinker with it. While it’s currently tested only in the Countdown game, Pan hopes this project will make reinforcement learning research more accessible.

Of course, it’s still early days. “One caveat, of course, is that it’s validated only in the Countdown task but not the general reasoning domain,” Pan admitted. But even with that limitation, the impact is clear: AI development doesn’t have to be expensive. With projects like TinyZero, affordable, open-source AI could be the future.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

For more daily updates, please visit our News Section.

Source | Via

The post UC Berkeley researchers managed to replicate DeepSeek AI for only $30 appeared first on Gizmochina.

The artificial intelligence industry has long been dominated by U.S.-based companies like OpenAI, Google, and Meta. However, the recent emergence of DeepSeek, a Chinese AI startup, has disrupted the global AI landscape.

DeepSeek’s latest model, DeepSeek-R1, has made headlines for its competitive performance, open-source nature, and cost-efficient development. As AI becomes an increasingly critical part of technological innovation, understanding how DeepSeek differentiates itself from other leading models is essential.

Open-Source Model vs. Proprietary AI

DeepSeek takes a different approach from U.S. AI giants like OpenAI by prioritizing open-source development. The release of its DeepSeek-R1 model under an MIT license gives researchers, developers, and businesses the freedom to access, modify, and deploy the technology.

OpenAI, on the other hand, has moved away from its initial open-source philosophy, keeping newer GPT models proprietary. This open-source approach provides greater transparency, fosters collaborative improvements, and lowers the barriers to AI adoption, making DeepSeek an attractive alternative to enterprises and independent developers who prefer not to be locked into proprietary ecosystems.

Model Parameters and Architecture

DeepSeek-R1 utilizes a Mixture-of-Experts (MoE) architecture, which significantly enhances computational efficiency. The model boasts 671 billion parameters, but due to its MoE setup, only 37 billion parameters are active at any given time. This enables it to achieve high-level performance while maintaining a lower computational cost than traditional transformer-based architectures.

In comparison, OpenAI’s GPT-4 is estimated to have around 1.8 trillion parameters, requiring significantly more computational power and leading to higher operational costs. DeepSeek’s approach allows it to optimize resources while maintaining accuracy and efficiency in text-based tasks.

Cost-Efficiency and Resource Utilization

DeepSeek has achieved AI performance on par with leading models like OpenAI’s GPT-4o, Meta’s Llama 3.1, and Anthropic’s Claude 3.5 Sonnet but at a fraction of the cost. The company claims to have trained DeepSeek-R1 with an estimated $5.6 million budget, significantly lower than the hundreds of millions of dollars spent by competitors.

This cost efficiency is largely due to its use of Mixture-of-Experts (MoE) architecture, which optimizes computation by activating only the most relevant parts of the model during each interaction.

DeepSeek reportedly trained its model using H800 GPUs, a slightly downgraded version of Nvidia’s high-performance H100 GPUs, which are restricted from export to China due to U.S. sanctions. Despite these hardware limitations, DeepSeek has demonstrated that it can develop highly competitive AI models with lower computational demands.

Reasoning and Coding Capabilities

DeepSeek’s AI models, particularly DeepSeek-R1, excel in technical tasks such as reasoning, coding, and mathematics. In third-party benchmarks, DeepSeek-R1 outperformed GPT-4o and other leading AI models in logical problem-solving, math computations, and code generation. According to reports, it scored 2,029 Elo on Codeforces, a coding competition platform, surpassing 96.3% of human participants.

It integrates chain-of-thought (CoT) reasoning, enabling it to break down complex problems into step-by-step solutions, a key area where OpenAI’s o1 model also focuses. While OpenAI’s ChatGPT remains superior in creative writing, conversational abilities, and nuanced human-like interactions, DeepSeek has carved out a niche as the preferred AI for developers, engineers, and researchers who need precise, logic-driven outputs.

Language and Market Focus

DeepSeek is uniquely positioned to serve both Chinese and English-speaking markets. Unlike OpenAI, which primarily optimizes for English-speaking users, DeepSeek is bilingual by design, excelling in both English and Chinese tasks. It has also outperformed American models in Chinese-language benchmarks, making it an attractive alternative for businesses and developers working in China and other Mandarin-speaking regions.

However, DeepSeek does exhibit language mixing tendencies, sometimes generating responses that blend English and Chinese even when prompts are given in a different language. This remains a limitation that the company continues to refine.

Pricing

The biggest advantage over its competitors is its affordable pricing model. DeepSeek-R1 offers API access at a rate of $0.14 per million tokens, making it significantly cheaper than OpenAI’s GPT-4o, which charges $7.50 per million tokens. This cost-effective pricing strategy has made DeepSeek an attractive option for developers and businesses looking for a high-performance AI model without the high operational expenses associated with competitors.

Security and Censorship Concerns

As a Chinese company, DeepSeek must comply with China’s strict internet regulations, which enforce content moderation around politically sensitive topics. Users have noticed that DeepSeek will not respond to queries about events like the Tiananmen Square massacre or China’s human rights record.

Conversely, OpenAI and other U.S. companies also implement content moderation, but they frame their policies around safety and ethical guidelines rather than government-imposed restrictions. This has raised concerns among privacy advocates and AI researchers who fear that DeepSeek could be leveraged for state-controlled narratives.

DeepSeek also got into trouble recently as its app was removed from Italian stores following a privacy investigation by Italy’s data regulator, Garante. Authorities are probing its data collection practices, raising concerns about compliance with EU privacy laws and potential security risks.

Market Impact and Industry Disruption

DeepSeek’s emergence has already shaken global AI markets. The launch of DeepSeek-R1 triggered a major selloff in AI-related stocks, with Nvidia losing $600 billion in market value over fears that lower-cost AI models could reduce the demand for expensive AI chips.

Following DeepSeek’s success, other Chinese AI firms such as Alibaba’s Qwen and ByteDance’s AI division rushed to update their own models, fueling a new wave of AI innovation in China. U.S. lawmakers and tech leaders, including President Donald Trump, have acknowledged DeepSeek as a potential threat to U.S. AI dominance.

Future Prospects and Industry Implications

DeepSeek is poised to continue growing, with ambitions to reach Artificial General Intelligence (AGI), a goal shared by OpenAI. However, its path will be shaped by factors such as international adoption, regulatory actions by U.S. policymakers, and continued AI innovation.

The AI arms race is intensifying, and DeepSeek’s rapid rise suggests that smaller, cost-efficient AI models may be the future rather than the largest, most expensive ones.

For more daily updates, please visit our News Section.

Tech enthusiast? Get the latest news first! Follow our Telegram channel and subscribe to our free newsletter for your daily tech fix!

The post What makes DeepSeek different from other AI models? appeared first on Gizmochina.

Heard of DeepSeek R1? Microsoft’s latest move is to bring distilled versions of DeepSeek R1 to Windows 11 Copilot+ PCs. It will allow developers to work on AI features that will run super efficiently on-device. Here’s everything you need to know about DeepSeek and its first AI model.

Microsoft has announced that it will soon offer developers the tools to build AI apps. For those unfamiliar, the DeepSeek R1-based AI features on Copilot+ PCs won’t require an internet connection to function. It also reduces concerns about data privacy as the user data will no longer need to be uploaded to the cloud for processing.

Here’s the rollout plan

Although the company hasn’t released an exact timeline yet, they have at least shared some details about it. The relatively new Snapdragon X-powered devices will be the first to run DeepSeek R1-based features, with Intel and AMD systems joining the party later. It’s not surprising though, as the Snapdragon X chips bring much more capable and power-efficient NPUs to consumer-grade devices.

Microsoft will start with “DeepSeek-R1-Distill-Qwen-1.5B”. The first wave of devices to get this upgrade are those powered by Qualcomm’s Snapdragon X Plus and X Elite chips. Think sleek laptops from brands like Asus, Lenovo, Dell, HP, and Acer. These devices already pack NPUs with 45 TOPS (that’s a lot of AI muscle), so they’re perfect for handling these new models.

Intel’s Core Ultra 200V devices are next in line, followed by AMD’s Ryzen AI 300 series. Interestingly, AMD’s already dropped some guides for DeepSeek R1 Distill, even for their older Ryzen 8040/7040 chips. Sure, those older chips don’t have the same NPU firepower as the newer ones, but hey, it’s nice to see AMD keeping things inclusive.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

For more daily updates, please visit our News Section.

The post Microsoft to bring DeepSeek R1 to Copilot+ PCs soon for efficient on-device AI processing appeared first on Gizmochina.

DeepSeek, the Chinese artificial intelligence startup, has found itself at the center of a regulatory storm in Italy after its app vanished from both Apple’s App Store and Google Play. The sudden removal follows a formal inquiry from Italy’s data protection authority, Garante, which is investigating how DeepSeek collects and processes user data. Concerns over data privacy and security have placed the AI company under intense scrutiny, echoing similar apprehensions raised in the United States and Australia.

Italian Privacy Regulator Investigates DeepSeek’s Data Practices

Italy’s privacy watchdog has given DeepSeek and its affiliated companies 20 days to disclose critical details about their data practices. Officials are demanding clarity on the nature of the personal data collected, its sources, intended usage, and whether it is stored on servers in China. The regulator has also asked how DeepSeek informs both registered and unregistered users about data processing, particularly in cases where information is acquired through web scraping.

Privacy concerns have been growing since DeepSeek’s recent meteoric rise. After launching an AI assistant that competes directly with OpenAI’s ChatGPT, the app surged to the top of download charts in multiple countries, triggering anxiety among competitors in the US tech industry. Meanwhile, American officials are assessing potential national security risks associated with a widely adopted Chinese AI model, and the US Navy has explicitly warned personnel against using DeepSeek.

Transparency over data handling remains a significant issue. The company’s privacy policy states that user data is stored on secure servers in China and may be shared with affiliated entities and service providers. However, Euroconsumers, a coalition of European consumer groups, has challenged the adequacy of these disclosures and questioned whether DeepSeek complies with the European Union’s General Data Protection Regulation (GDPR).

Italy’s scrutiny of DeepSeek is not unprecedented. In 2023, the country briefly banned ChatGPT over concerns about user data protection. OpenAI responded to these concerns by implementing several changes to the platform, including greater transparency about data processing, opt-out rights for users, and age verification measures to protect children under 13. These adjustments led to the chatbot being reinstated.

As DeepSeek gains traction globally, its regulatory challenges are mounting. The company must respond to the Italian watchdog’s questions by February 17, a deadline that could determine the fate of its operations in the European market. If authorities find violations of privacy laws, DeepSeek could face substantial penalties or restrictions, potentially setting a precedent for future scrutiny of AI applications developed outside of the West.

Check out our in-depth article on DeepSeek, China’s AI breakthrough that’s making waves and challenging the U.S. tech landscape.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories!

(Via)

The post DeepSeek app removed from Italian app stores over privacy concerns appeared first on Gizmochina.

DeepSeek, a Chinese AI company, is making big waves in artificial intelligence. On January 20, 2024, they launched DeepSeek R1, a powerful language model. This model is open-source and affordable, making it a strong competitor to OpenAI’s top models. DeepSeek’s success is challenging the US’s lead in AI development and sending shockwaves through the global tech industry.

What is DeepSeek R1?

DeepSeek R1 is a new kind of AI model built to think and solve problems more like humans. Instead of just using huge amounts of data, it learns in real time, recognizes patterns, and improves on its own. This makes it smarter and more independent, moving AI closer to real human-like reasoning.

Key Features of DeepSeek R1

DeepSeek R1’s open-source nature and focus on cost-efficiency offer a stark contrast to Western AI giants, positioning it as a game-changer in the field of artificial intelligence.

The technology behind DeepSeek R1

DeepSeek R1’s architecture combines efficiency and intelligence, setting a new benchmark for AI performance. While traditional AI models require massive computational power, DeepSeek optimizes its neural networks to achieve higher accuracy with fewer resources. This efficiency is crucial for industries like healthcare, finance, cybersecurity, and autonomous systems, where real-time decision-making is essential.

By utilizing a self-learning framework, DeepSeek reduces reliance on extensive manual programming, making it more adaptable across various industries.

DeepSeek’s Open-Source model gives it an edge over ChatGPT

A key feature of DeepSeek R1 is that it is completely open-source. Unlike OpenAI’s models like ChatGPT, which have restrictions, DeepSeek R1’s code is free for anyone to use and develop. This open approach will help AI grow faster by allowing developers worldwide to improve and build on it. 

Open-source AI is also important because it gives everyone access to advanced technology, not just big companies. It helps developers improve and customize AI, making it better for everyone. Being open also builds trust and encourages teamwork, leading to faster progress in AI.

DeepSeek’s disruptive AI development strategy

DeepSeek’s strategy of developing powerful AI models with limited resources sets it apart from its competitors. The R1 model, with only 671 billion parameters, is significantly smaller than those of its rivals. 

Only 37 billion parameters are activated during usage, optimizing computational resources. The model was trained with just $5.6 million, far less than the multi-billion-dollar investments made by OpenAI and Google.

How does this impact the AI Market?

The financial impact of DeepSeek’s emergence has been significant. Following the announcement, the Nasdaq Composite Index dropped over 3%, with major U.S. tech firms, including Nvidia, Oracle, and Dell, suffering substantial losses. The world’s 500 richest individuals lost a combined $108 billion, with Nvidia CEO Jensen Huang losing $20.1 billion in a single day.

DeepSeek has gained attention for creating powerful AI models at a much lower cost than big tech companies in Silicon Valley. While Western firms rely on expensive Nvidia GPUs, DeepSeek found a more efficient way to train its AI. Before U.S. export rules tightened, the company collected thousands of Nvidia GPUs, giving it a big advantage in computing power.

Who Created DeepSeek? 

Much of DeepSeek’s success can be attributed to its visionary founder, Liang Wenfeng. Born in 1985, Liang studied information engineering at Zhejiang University before starting DeepSeek in 2023. He has also established High-Flyer Quant, one of China’s largest AI-driven hedge funds. Liang’s emphasis on open AI research and original innovation distinguishes DeepSeek from many other Chinese tech firms, which often focus on adapting existing technologies.

Startup culture is at the heart of this innovation 

DeepSeek operates with a start-up mentality. Its team consists of highly skilled graduates from top Chinese universities, including Tsinghua and Peking. The company follows a flat organizational structure, fostering flexibility and rapid innovation. The development of the DeepSeek R1 model involved around 200 researchers, highlighting the company’s collaborative approach to AI development.

AI investments in Silicon Valley are now under huge pressure 

DeepSeek’s success challenges the traditional belief that cutting-edge AI development requires massive financial investments. With a reported development cost of just $5.6 million, DeepSeek has shown that efficient AI training methods can rival billion-dollar projects from Western tech giants. This has led to investor skepticism about Silicon Valley’s high-spending AI strategies, raising questions about whether U.S. firms are overinvesting in AI development. Billions of dollars vanished from the market as investors panicked over DeepSeek.

The Global AI Race: U.S. vs. China

As the U.S. and China continue their battle for AI supremacy, DeepSeek’s rise shifts the dynamics of this global competition.

Despite U.S. sanctions on high-performance AI chips, China is countering by developing its own AI hardware and optimizing models to run on lower-cost chips. This evolving competition is reshaping global AI policies, with both nations striving for dominance in next-generation intelligence systems.

How is the U.S. responding to DeepSeek’s growing influence in AI?

The U.S. government has taken notice of DeepSeek’s disruptive impact. White House press secretary Karoline Leavitt confirmed that U.S. officials are reviewing the national security implications of the Chinese AI model. The National Security Council (NSC) is evaluating potential threats posed by DeepSeek’s AI technology. Concerns have been raised about the potential distillation techniques used by DeepSeek, leading to discussions about stricter measures to prevent unauthorized learning from U.S.-based AI technologies.

US policy shifts and the future of AI regulations

DeepSeek’s rise has forced the U.S. government to reconsider its AI and semiconductor policies. Analysts see DeepSeek’s success as a “Sputnik moment” for the AI race, similar to the Soviet Union’s launch of the first satellite in 1957. U.S. policymakers may shift their focus from hardware bans to regulating AI model behavior and access to training data in the future.

Is this the start of a new age in AI?

DeepSeek’s emergence marks a turning point in the global AI competition. It has demonstrated that China’s private sector can lead in AI innovation, challenging U.S. dominance in the field. As the U.S. reassesses its AI policies and China strengthens its tech ecosystem, DeepSeek is poised to play a pivotal role in shaping the future of artificial intelligence. 

Whether this leads to increased competition or collaboration remains to be seen, but one thing is clear, DeepSeek has cemented itself as a game-changer in the AI landscape.

Read More:

• China’s DeepSeek AI Assistant is now the top free app for iPhones
• Amap unveils China’s first English map: A game-changer for foreign tourists
• China’s first AI railway robot is here and it’s a game-changer

The post All you need to know about DeepSeek: China’s AI game changer and a wake-up call for US appeared first on Gizmochina.

DeepSeek, the brand-new disruptive AI from a Chinese startup, is showing no signs of stopping. We recently talked about DeepSeek AI as it challenges OpenAI’s dominance with ChatGPT. Now, in just a short time, the DeepSake AI assistant app has become the top free app on iPhones.

DeepSeek AI Assistant Top App on Apple’s App Store on iOS

As of right now, the Chinese AI assistant from DeepSeek is the top rated free app on Apple’s Apple Store in the US. It is currently the top app for iPhones, beating out rivals like OpenAI’s ChatGPT. For those unaware, DeepSeek is making waves for being an open-source platform, which is based on the DeepSeek V3 model. The artificial intelligence apparently requires far less computing power than its competitors as well.

While it is still disputed, the people behind DeepSeek claim that it was developed for under 6 million US Dollars. Despite its relatively cost efficient development, it promises to offer performance that can match Claude 3.5, GPT-4o and others. Similar to other AI platforms, DeepSeek also offers functions like ChatGPT, which can help content creators and be useful in research. It is currently available as an app, API, and on the web, making it easily accessible to all.

The first DeepSeek-R1 was rolled out under an MIT license, so it is free to be used commercially without any restrictions. DeepSeek’s open-source approach also challenges various closed-source models developed by major tech brands. This shift towards greater transparency and accessibility, allows a wider range of individuals and organizations to contribute to its development and benefit from its potential.

For more daily updates, please visit our News Section.

Don’t miss a thing! Join our Telegram community for instant updates and grab our free daily newsletter for the best tech stories!

The post China’s DeepSeek AI Assistant is now the top free app for iPhones appeared first on Gizmochina.

A new player, DeepSeek, a Chinese AI startup, has shaken up Silicon Valley with its cost-efficient language model, DeepSeek-R1, rivaling OpenAI’s ChatGPT. Despite US export controls on advanced AI chips, the company has achieved breakthroughs through innovative strategies, prioritizing efficiency and performance. This development is reshaping the AI landscape—read on to learn more.

DeepSeek’s V3 and R1 AI language models

Unlike many Western AI companies that focus on scaling up by acquiring vast amounts of computing power, DeepSeek has taken a different approach. Faced with US export controls on advanced chips, the company focused on optimizing software and algorithms to maximize efficiency.

DeepSeek offers two advanced AI models: DeepSeek-V3, designed for a wide range of applications, and DeepSeek-R1, a cost-effective alternative to ChatGPT.

DeepSeek-V3, an advanced AI language model, is designed for a broad spectrum of applications, including natural language processing, customer service, education, and healthcare. Optimized for understanding the Chinese language and its cultural context, DeepSeek-V3 also supports global use cases. The model is focused on delivering high performance while being cost-effective and efficient, making it a versatile tool for various industries, particularly within the Chinese market but adaptable for international markets as well.

DeepSeek-R1, another model from DeepSeek, offers performance comparable to OpenAI’s ChatGPT at a significantly lower cost. Despite facing challenges such as US export controls on advanced AI chips, the model maintains high-quality results through efficiency and innovative approaches. Its primary goal is to serve as a cost-effective alternative to other AI models like ChatGPT, positioning DeepSeek as a competitive player in the global AI market. With a focus on overcoming resource limitations, DeepSeek-R1 embodies the company’s commitment to innovation and performance at scale.

DeepSeek’s founder, Liang Wenfeng, a former quant hedge fund manager, has assembled a team of young, ambitious researchers from China’s top universities, providing them with ample resources and freedom to explore unconventional ideas. This approach has led to the development of groundbreaking techniques like Multi-head Latent Attention (MLA) and Mixture-of-Experts, which significantly reduce the computational resources required to train their models.

DeepSeek vs. ChatGPT: A short comparison 

DeepSeek-V3 and ChatGPT are both advanced AI models, but they differ in key aspects. DeepSeek-V3 is optimized for Chinese language understanding and cultural context, while also supporting global applications. It is particularly tailored for industries like education, healthcare, and customer service, with a strong focus on the Chinese market. In contrast, ChatGPT, developed by OpenAI, is trained on a globally diverse dataset with a stronger emphasis on English and Western contexts, making it widely used for general-purpose tasks, creative writing, coding, and more.

Both models are highly capable, but their performance may vary depending on the task and language, with DeepSeek-V3 potentially excelling in Chinese-specific tasks and ChatGPT performing better in English-heavy or globally diverse scenarios. Additionally, while both models adhere to strict ethical guidelines, their alignment may differ slightly based on regional regulations and cultural norms.

DeepSeek’s Global Impact

DeepSeek’s commitment to open-source development has garnered praise from the international AI community. By making its models freely available, DeepSeek is fostering collaboration and accelerating AI research worldwide. This is particularly significant for researchers and developers in the Global South who may have limited access to expensive proprietary models.

DeepSeek’s open-source approach also challenges the current trend of closed-source models developed by major tech companies. This shift towards greater transparency and accessibility could democratize AI technology, allowing a wider range of individuals and organizations to contribute to its development and benefit from its potential.

DeepSeek’s models, including the powerful DeepSeek-R1, are available globally using its URL: https://chat.deepseek.com/. While the company is based in China, its open-source approach allows anyone, regardless of location, to access and utilize its technology. This has significant implications for the future of AI development, as it allows for a more diverse range of contributors and accelerates the pace of innovation.

For more daily updates, please visit our News Section.

Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories! 

The post DeepSeek: The Chinese AI startup challenging OpenAI’s dominance appeared first on Gizmochina.